Skip to content
Theme
OnDevice Tools

GadgetsLaptops & PCs

Intel details Diamond Rapids Xeon chips and GPU D-series for supercomputing

Intel has outlined its roadmap for high-performance computing, covering new Xeon processors with up to 256 cores, AI accelerators with large memory pools, and work on optical networking and quantum computing.

Illustration of a server rack with Intel Xeon processors powering a supercomputing data centre

Key takeaways

  • Intel's Diamond Rapids server processors will support up to 256 performance cores, 1.6 TB/s of memory bandwidth and 1.28 GB of last-level cache.
  • The GPU D-series accelerators will offer up to 480GB of LPDDR5X memory and formats from FP4 to FP64, with air cooling and PCIe connectivity for easy installation.
  • Intel is keeping its software tools such as oneAPI and Intel MPI aligned with new hardware so research institutions can keep existing applications.
  • The company is also pursuing silicon photonics for data centre networking and has proposed acquiring Classiq to pair with its spin-qubit quantum research.
In this article
  1. What Intel announced
  2. Diamond Rapids raises core counts and memory speed
  3. GPU D-series targets memory-hungry AI work
  4. Software compatibility remains central
  5. Optical networking and quantum research

What Intel announced

Intel has reaffirmed its commitment to high-performance computing, the field of supercomputing used for scientific research and large-scale AI, according to TechRadar. The company is working on new Xeon server processors, dedicated AI accelerators, optical networking technology and quantum computing research.

The aim is to let research institutions adopt faster hardware while keeping their existing software and infrastructure.

Diamond Rapids raises core counts and memory speed

The centrepiece of the roadmap is Diamond Rapids, Intel's next-generation server processor family. Configurations will scale up to 256 performance cores, with memory bandwidth reaching 1.6 TB/s and last-level cache of 1.28 GB, according to TechRadar.

The chips will also carry new instruction set features aimed at scientific and AI workloads. These include Intel Advanced Performance Extensions, the AVX 10.2 vector instruction set, and enhanced matrix extensions that accelerate demanding calculations in hardware.

Intel positions Diamond Rapids as a successor to its Xeon 6 processors, which already support intensive simulation and mathematical modelling. The company says the new architecture extends rather than replaces that lineage.

GPU D-series targets memory-hungry AI work

Illustration: privacy, AI, film and TV, graphics card and the labels Diamond Rapids, Intel, Xeon

Alongside the processors, Intel is developing its GPU D-series accelerators for research organisations that need large accelerator memory together with AI processing and double-precision maths. Each accelerator will support up to 480GB of LPDDR5X memory, a fast type of memory mounted close to the chip.

The supported numerical formats span a wide range, from compact FP4 and MXFP4 formats used in AI inference up to FP64, the high-precision format common in scientific computing.

Intel has designed the cards for conventional PCIe slots and air cooling, so they can be installed in existing facilities without special cooling systems. Larger onboard memory could also cut the repeated transfers of data between processors and accelerators when applications handle very large datasets or models.

Software compatibility remains central

Intel argues that faster processors are of limited value if researchers must rewrite established applications. Its software portfolio therefore includes Intel MPI, Fortran compilers, optimisation tools and oneAPI, which let parallel applications run across different hardware configurations.

The company plans to have software support ready when new hardware launches, so widely used AI frameworks and inference environments work from day one. This also enables hybrid workflows in which standard processors coordinate applications while accelerators handle the heavy AI computations.

Optical networking and quantum research

Intel is also investing in silicon photonics, a technology that uses light instead of electrical signals to move data. It could reduce energy use and delays in the connections between processors, memory and accelerators across very large supercomputing installations.

On quantum computing, Intel has proposed to acquire Classiq, whose software platform simplifies quantum programming. The plan is to combine that automated algorithm development with Intel's silicon spin-qubit technology, potentially letting developers familiar with conventional supercomputing explore quantum applications without managing complex hardware directly.

Sources

  1. 'Intel’s commitment to the HPC community is grounded in a pragmatic reality': Intel affirms its backing for the next generation of high-powered computing to push the boundaries of AITechRadar

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.